Hacker Newsnew | past | comments | ask | show | jobs | submit | joshrw's commentslogin

Chinese models optimize for benchmarks and do poorly in real-world tasks


Not my experience at all, I have written about comparing DS4 vs Opus 4.8 on 16 real life work tasks on multiple posts.

Also, every single lab does RL on benchmarks, which is why Opus 4.6 was the last truly great assistant, after it, all models tend to drift into implementation asap.


Hi, author here, can you link? I would love to read about this.



It wasn’t predictable that he would start a war.

He presented himself as the anti-war candidate and then betrayed his electorate.


From outside the US, nothing about him has really been predictable, except that he consistently lies about everything.

The only predictable thing about this American presidency is the total chaos that has been inflicted on the world by millions of intellectually degenerate Americans. This is what the world sees. Almost nobody is making excuses for him, everybody is trying to move on without the US.

So yeah, in that sense another war is simultaneously an unpredictable outcome and an unsurprising outcome.


The real question is should you have trusted his manifesto after the last time?


During his first term, his continuous threats to north Korea were enough to force congress to restrict presidential powers (the whole fire and fury stuff), and that's not counting Venezuela, Syria and others.


It was evident to anyone aware and paying attention.


Since January 6th it was clear that he would be willing to secure his power in any way. The most obvious way in america is to start a war.

And even if noone could predict that - the worst part is all the Americans sitting by, twiddling thumbs after seeing Trump making the whole world worse. You are the loudest "democracy" on the planet, but noone demands the necessary accountability from the dear leader. Democracy does not only happen on election day.


He presented himself as someone who has utter disregard for any form of coherence and people actually argued that that's what made him palatable to the masses. I don't think that someone who voted for him can in good faith claim they didn't know he would betray his promises. All they wanted Is for him to own the libs


How many wars and military actions did he start in his first term? How many times did he do the opposite of what he said he did? The specific case is unpredictable, the general pattern isn't.


I agree with this. What was certain and predictable was that Harris was a war candidate (the Democrats presided over the first year of genocide and were intending to continue).

MAGA was going to be bad in some way, but the Democrats were very bad in very known ways, so voting for whoever was at least saying they were against war was the rational decision for anyone who cared about not having genociders at the helm of the state.


To give you the benefit of the doubt at best it would be like 5/10 vs 6/10. There wasn’t ever any difference that Soto’s have ever recruited into any decision. The only reason someone would have thought Harris was a war candidate would have been because rogan or similar said it.


Both Harris and Trump were identical in that respect, so it shouldn't have swayed your decision.


Harris was a known to be bad with respect to the war issue, became very obviously bad when she started campaigning with Liz Cheney. Trump was less known bad, so that difference should certainly have swayed your decision.


Trump had already tried to overthrow the previous election. He could only have been thought less bad if you weren't paying attention.


If you define paying attention as funneling MSNBC straight into your brain 24/7, then sure.


Greater variance, same mean


>MAGA was going to be bad in some way, but the Democrats were very bad in very known ways, so voting for whoever was at least saying they were against war was the rational decision for anyone who cared about not having genociders at the helm of the state.

This implies that Donald Trump/Republicans weren't pro-Zionist and didn't support Israel which is obviously untrue.

The truth is there was no "non-genocide" candidate between the two parties and pretending otherwise - particularly by voting for Trump - was simply self-delusion.


Both sides are against the genocide of Israel, thank God - and against the genocide of Arabs.

(Unlike the meme-puppets who now believe that some genocides are better than other genocides).


If you paid attention to his actions while in office, rather than a selection of his words while on the campaign trail, it was pretty clear he wasn't in the least bit interested in being a "peace president;" he just wanted the Nobel Peace Price because Obama got one.


The GPT-4 demo. Taking a screenshot of handwritten instructions to build a website, along with a drawing of what the website should look like. Then ChatGPT spit out a working prototype.

Also the live video mode demo later that year.

Then the agentic coding breakthrough in Nov/Dec 2025.


That’s not what that quote means. It means even if your portfolio is down you should keep buying.


That's what they meant by "too literally."


No, the "quote" (more like anonymous anecdote) is quite literally about blood running in the streets presenting an opportunity to buy assets at reduced cost.



“Bug”. More like a “bugdoor”


Terrible article. The real Satoshi is Nick Szabo and no one else is even close. Hal Finney, Wei Dai, etc. New York Times’ quality has really gone downhill.


Hello, have you heard of the Snowden revelations? What OP was referring to are called bugdoors.


I'm very concerned about bugdoors and very grateful to Snowden, but I don't remember a specific example of a software bugdoor that was disclosed there or identified as such as a result of his revelations. Do you have an example? I don't think the Dual-EC DRBG counts here.


This happened later than Snowden, but is an example of an unsettling revelation.

A bug has existed for many years in Apple devices, until a few years ago, when it has been discovered accidentally by some victims, which has forced Apple to fix it, after several CVEs where assigned to it and associated software bugs.

The bug consisted that some secret test registers, which allowed a complete bypass of all memory protection, were left accessible after production. Thus knowledgeable attackers could take control remotely of an iPhone, for many years, in a completely undetectable way, by sending an invisible message, which then exploited some bugs in Apple system libraries to gain privileged access to the secret test registers, which were then used for complete access to any hardware, including stored files, video camera and microphone.

This backdoor was discovered only because some victims became suspicious due to unexpectedly high Internet traffic originating from their iPhone, which was recorded by an external firewall.

This was discussed on HN after its discovery.

It is hard to believe that such a mistake like forgetting to disable the test registers after production could have happened and it also would have never been discovered for many years, without some Apple insider intentionally doing it.

Moreover, the unknown attackers who have exploited the backdoor for many years had complete knowledge about the secret test registers, which is likely to have been provided by an Apple insider, perhaps the same who has ensured that they remain accessible.

Hopefully, the backdoor has been created only by some lower-rank employee, and it was not created with the knowledge of the management, due to some request from a TLA. It is unknown whether the backdoor has been open in all Apple devices, or only in those sold in certain markets.

When the backdoor was discovered, it was used to spy on some Russians, so some US agency or one from Israel were among the possible exploiters of it (this was before the current war).


Doublespeak


Then it needs to do context compacting, otherwise the results become garbage


Happening very often lately


and we all know why


Because they're moving it to Azure and doing it far too quickly, not taking care to avoid availability issues


It wasn't the migration to Azure that completely borked their PR UI.


Could be.

Or could be that the recent 12 months of 100x increase in code and activity is more than they had planned for when they last did capacity planning.

Vibe-coders, many of them here, often boast about the insane amount of KLoC/hour they can generate and merge.


I've seen this take in another GitHub thread, but are there any stats confirming this? As far as I know a lot of Github stats are publicly available, and can be queried via Clickhouse.


There may be other problems but as someone who's somehow ended up integrating Git into a service twice in my career without even trying that hard to find a reason (it turns out it's weirdly handy in quite a few situations, god I wish it were implemented as a library and not a pile of Perl and shit, and yes I know about libgit2) and has looked into some of Git's and Gitlab's posts about their architectures over the years though the lens of having fought a few of the same beasts, an Azure migration was very obviously going to make things worse.


yeah, ai slop rush

everyone builds off vibes and moves fast! like no, if you are a mature company you don't need to move fast, in fact you need to move slow

the only thing that can kill e.g. github is if they move fast and break things like they do recently


Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: